Rumen Digital Twin

Rumen Digital Twin Project - Alliance Bioversity International and CIAT

The Rumen Digital Twin is an AI-powered foundation model that integrates data on ruminants, diets, environments, genetics and rumen microbiomes to generate realistic virtual animal populations. It will help researchers and organizations investigate how these factors interact and predict where feed additives and other interventions can be most effective.

Project Name (full): Rumen Digital Twin for sustainable livestock production

Start and end year: 2025-2027

Region and Countries: Americas

Funder: Bezos Earth Fund’s AI for Climate and Nature Grand 

Partners: Yale University and Biomedit

Brief description

Livestock research generates large amounts of data on animal genetics, diets, microbiomes, methane emissions, productivity and environmental conditions. Yet results from experiments can vary substantially across studies because animals, feeds, climates and experimental conditions differ. At the same time, the complex interactions among these factors in the rumen are not fully understood, making it difficult to explain why the same feed or intervention may produce different outcomes across animals, countries or production systems.

The project is developing a Rumen Digital Twin (RDT) foundation model that brings these diverse datasets together to learn the relationships among the factors that shape ruminant performance and methane emissions. Rather than attempting to simulate every biological process inside the rumen, the model learns from measurable data to create realistic representations of animals and their interactions.

A key innovation is its generative capacity: researchers could use the model to create virtual cohorts of animals with specific characteristics and test hypotheses in silico before conducting physical experiments. This could help identify promising research directions, improve predictions and potentially reduce the number of animal experiments needed in early-stage research. By providing a shared foundation for research, the project aims to turn fragmented knowledge into a reusable resource for understanding and improving livestock systems while potentially reducing the need for some early-stage animal experiments.

Rumen Digital Twin Project - Alliance Bioversity International and CIAT - Image 2

Key activities

  • Building a harmonized global dataset: The team is identifying, collecting and processing publicly available data on ruminant microbiomes, host characteristics and diets to build a comprehensive, harmonized database that integrates phenotypic and microbiome data across studies, species, breeds and regions. This will be complemented by data generated by the project and global initiatives, covering more than 20,000 ruminants across 30+ countries. As a result, over 10,000 rumen microbiome profiles crosslinked with diet compositions, and host properties such as methane emissions, genetic markers, and performance parameters will be generated.
  • Developing and validating the foundation model: We will train our model by adapting an approach we recently published. The model will integrate microbiome and host information into a shared latent space (an encoded representation). It will be continuously retrained as new data becomes available and tested against data withheld from training to assess its ability to reproduce observed rumen characteristics from partial data and capture relationships across different data types.
  • Model fine-tuning for predicting intervention efficacy: To predict the characteristics of animals in which specific feed additives or diet interventions would be most effective at reducing methane while maintaining or increasing productivity the model will be fine-tuned using data from 3–5 feed or diet interventions, including anti-methanogenic tropical forages from the Low-Methane Forages project. Predictions for 2–3 interventions will be tested in targeted animal experiments, with results used to refine the models.
  • Engaging future users: Throughout the project, researchers and organizations will help explore potential applications and co-develop Rumen Digital Twin through providing information and datasets. The final product will be a web application that will allow users to explore the model and generate virtual cohorts of ruminants.
Rumen Digital Twin - Alliance Bioversity International - CIAT
Rumen Digital Twin Project - Alliance Bioversity International and CIAT - Image 1

Relevant blogs, stories, news coverage

Relevant publication

Other Project Members: Mauricio Peñuela, Germán Plata, Jacob Shields, Dwi Susanti, Ashley Hall, Sandra Calderon, Arvind Kumar, Purushottam Dixit

Project leaders

Project members